Cancelable fingerprint template design with randomized non-negative least squares

Abstract

Although biometrics is considered more competent than password-based or token-based approach in identity management, biometric templates are vulnerable to adversary attacks that may lead to irreversible identity loss. One of the promising remedies for biometric template protection is cancelable biometrics. In this paper, a novel binary cancelable fingerprint template design based on Partial Local Structure (PLS) descriptor and Permutated Randomized Non-Negative Least Square (PR-NNLS) is proposed. The PLS descriptor is an alignment-free minutia descriptor, which is conceived to be coupled with the PR-NNLS to derive a binary protected fingerprint template that satisfies non-invertibility, unlinkability, cancelability and performance criteria. The PR-NNLS formulation is unique in such a way that the noninvertible transformation is applied to the PLS descriptor dictionary instead of applying it to the minutiae descriptor, which often invites performance deterioration. The evaluations have been carried out with five subsets from FVC 2002 and 2004 databases where the proposed method is attested to fulfill the aforementioned four template protection criteria. We also analyze four privacy and security attacks targeted to cancelable biometrics. (C) 2019 Published by Elsevier Ltd.